Developing projects with AI

[Vod]

Thanks for sharing the guide! It’s definitely worth a shot on a machine with a GPU. The experience difference between using your CPU and even a 3060 would be huge!

YW, but it's almost a year old now.  I moved from DS to ChatGPT and now I'm with Gemini and Antigravity (thx to @TryNinja for the hookup!).  I think my next step will be a dedicated server for a LLLM.  (Local Large Language Model) so I can finally move away from Google Assistant playing the wrong music at night and ruining my mood.  :/

release date jan 14 means you are going to be busy over christmas lol

Almost done, actually!  I can release a BETA before xmas, but who wants to code when spiced rum and eggnog is socially acceptable?  I think the natural release date is Jan 1, 2026.

[Vod]

My AI project is complete, and using Antigravity it took me 14 days in total!  The design is gorgeous, there are no bugs, and it is lighting fast.  This was my test project, and now I'm going to start work on a bitcoin browser game.   It's based off an actual fun game I use to play, and it will reward bitcoin as prizes. 

There is a lot of negativity recently around AI, for good reason, so I want to explain how I am doing something positive with AI.

First, I have a background as a coder.  I started in 1983 so I know what I am doing, but the programming languages, best practices and open source tools change so rapidly I can no longer keep up.  I also was never a designer but I rely on design for inspiration, so I had to pay someone to make every image and change them as my ideas changed.   These would probably not be large issues if I didn't have a brain injury, but with my poor typing dexterity in my left hand, I can make typing mistakes constantly, leading to slow development and frustrating troubleshooting.  For any programmer, using AI to write a common subroutine would be no different than maintaining your own script repository, but they would frown upon AI writing original code. 

Programming this extension took me about the same time it would have if I had coded it myself in my younger days (minus the design).  AI is great for "slop", or generating existing code but if you ask it for something new, it can really go wild in its logic.  You must be precise in prompts, plan for future features yourself and make sure AI doesn't remove the novel code you wrote.    I spent hours revising specific features, but I did not code those revisions myself. 

To me, that's where AI turns from a garbage generator to a lifesaver.  You won't see me creating forum content with AI (wearing a sig would make that fraud) but I am using AI to help me write the exact specific code some systems require.  Things I can no longer do myself.

[IsraelK]

I wanted to provide an update on my AI project experience.

When I started two months ago, I would have told you I could complete my project before I went to sleep - Deepseek was that good at doing exactly what I wanted...  Over the last few weeks I've struggled to make much progress just like when I was coding manually.  When I first told it what I wanted, it designed it perfectly.   Then I started to add features and that is where the problem started.

DS would write a function a different way than it wrote for me a day before (example:  passing a parameter in post body vs querystring) and then my code would break.  Rather than fixing what I knew was the issue, I wanted DS to recognize it.  It did, but only after I provided fresh affected code and/or table definitions.   Since I used DS to code things I had no idea of (event handlers on browser clicks for example), it would tell me I was missing this and add it to that, but then later it may tell me to add the same this to a different that, duplicating functions and leading to troubleshooting anxiety.   Also, every now and then DS will give me code that contains a missing closing tag...

I wanted AI to code this entire project based on my input.  I continue to believe that AI is great for mundane tasks like syntax or logic checking, but it cannot be used as a substitute for overall project development.   If I wasn't a coder before, I would have no idea what to ask or tell DS.    Those of you who are waiting on some features I promised; I still feel I can complete it by the end pf this month.

I can definitely relate to this. AI is impressive at getting an initial design or feature off the ground, but once a project grows, consistency and architectural awareness become the real challenge. Another issue I’ve noticed is that AI doesn’t retain a true mental model of the entire codebase, so small implementation differences can quietly compound into bigger problems. Used as a productivity aid rather than a replacement for engineering judgment, it’s still incredibly valuable.

[Vod]

I can definitely relate to this. AI is impressive at getting an initial design or feature off the ground, but once a project grows, consistency and architectural awareness become the real challenge. Another issue I’ve noticed is that AI doesn’t retain a true mental model of the entire codebase, so small implementation differences can quietly compound into bigger problems. Used as a productivity aid rather than a replacement for engineering judgment, it’s still incredibly valuable.

Actually, antigravity is incredible at setting boundaries.  I connected it to my github account and my gemini subscription and it can modify just a small section without touching any other code.  I LOVE it.  If I hadn't put my development trust into a crappy web host, I might had of completed my project in just a week.

ShadowPulse has been released!
https://bitcointalk.org/index.php?topic=5569385

My next project will be a bitcoin game. 

[Vod]

One of the larger programming Q&A sites has been killed by AI.

Stack Overflow Went From 200,000 Monthly Questions To Nearly Zero

The last programming question I asked of a human was a month or so ago when I asked TryNinja "What do you use to program?".   Since then I have rebuilt my home media server and improved wifi efficiency, as well as programming.  Is there anyone who still thinks AI is just a fad?

[dannyatwork8]

One of the larger programming Q&A sites has been killed by AI.

Stack Overflow Went From 200,000 Monthly Questions To Nearly Zero

The last programming question I asked of a human was a month or so ago when I asked TryNinja "What do you use to program?".   Since then I have rebuilt my home media server and improved wifi efficiency, as well as programming.  Is there anyone who still thinks AI is just a fad?

Pretty sure it had to do with their utter nonsensical gatekeeping and censorship rules instead of just GPTs killing them. Also, LLMs relied heavily on human curation on Stack for its training. How is it going to learn any new data without it?

[Vod]

Also, LLMs relied heavily on human curation on Stack for its training. How is it going to learn any new data without it?

A year ago, they did.   Now new LLMs are training with curated, organized copies of that original data.  They are able to learn new things based on the new classifications and connections with existing data.  Then specialized LLMs fire up on specific categories of information, becoming an expert in just a single field (so it does not become a threat to its creator) and fill in the remaining information from conversations with users.

Just like BPIP opened the door to the value of Data Dictionaries (aka metadata), new AIs will discover patterns and inventions without human imagination.

[programmer3666]

One of the larger programming Q&A sites has been killed by AI.

Stack Overflow Went From 200,000 Monthly Questions To Nearly Zero

The last programming question I asked of a human was a month or so ago when I asked TryNinja "What do you use to program?".   Since then I have rebuilt my home media server and improved wifi efficiency, as well as programming.  Is there anyone who still thinks AI is just a fad?

Pretty sure it had to do with their utter nonsensical gatekeeping and censorship rules instead of just GPTs killing them. Also, LLMs relied heavily on human curation on Stack for its training. How is it going to learn any new data without it?

I can attest to this!! I am a programmer and I can tell you for a fact that it's been ages since I last visited Stack Overflow to look for a solution to any of my coding problems. AI has really outdone most of these community driven platforms. I am sure there are still some fundamental issues you can come across and you will have to take a look at them for a quick fix. Personally, I use ChatGPT to get some quick solutions to my coding problems, but I will try out Deep Seek and see how it goes.

[Free Market Capitalist]

Is there anyone who still thinks AI is just a fad?

No, especially if we consider that it is still in a very primitive state. I have been reading the thread, and although I know nothing about programming or code, I see that you are doing what any rational and open-minded person does today with AI.

You test an AI, see how it works, where it tends to fail, refine the inputs, compare it with other AIs, etc.

I am especially glad that AI provides you with leverage where you have physical difficulties.

I'll keep following your journey.

[Vod]

I'll keep following your journey.

You're gonna have to stalk me, brother.    I'm closing this thread.

I'm still using AI to assist me rather then replace me and I am getting sick of all the constant AI ads I see on IPTV.    The time to get involved in the revolution was a couple years ago.  None of these commercial products (built on the LLMs) are superior to them but they obfuscate the creepy overreach these models are achieving.  I'm also closing this thread because the line between reality and AI is not something I want to encourage.

[Vod]

I don't know why I locked this thread, stupid to shut down discussion on a continuing issue.   Sorry

In the past I've bugged TryNinja about his AI tools and I got great results.   But now I'm hitting limits in what I want to do and I'm wondering what the current status on AI tools is?

I'm currently using Kimi Claw for automation, but it always just "stops" for no reason and I have to poke it to continue.  If you use an AI integrated with an IDE, I'd love to hear your experience!

[DYING_S0UL]

I opened up DeepSeek and typed something like "Using this page (https://bitcointalk.org/index.php?action=profile;u=30747) can you identify all values and put them in a C# datatable?" and within a minute it spit out code that took me weeks to first write.       I have been using it for what all programmers hate - error handling.    And with the incredible context Deepseek manages, I don't need to issue the same instructions over and over.  I can write "How about this code?" and it will apply every necessary check for nulls, etc., as long as I asked it to do so previously in the same chat.    Being in Alberta, I can use it easily until around 2200, then the server starts getting busy.

So I installed Deepseek on my local server (HP DL380) with 64GB RAM and Xeon processors.   Downloaded the 90GB database in under 5 mins* (lol) and it installed fast!   But there is no video card in this server.   I asked it "How can I reduce cat allergens?" - it took 17 mins to think and then spit out a word about every two seconds.  Not usable at all.    That was the full R1:70b.   So I tried the limited R1:8b, got a notice there was a new version and asked "What version are you?" and it still took almost five mins thinking.

* When I first got into BBSes, I could download 1MB an hour.   Now I can download 100MB a second (a 360,000x increase in 30 years).

If you are interested in trying DeepSeek on a computer with a video card, here is a straightforward guide for running the AI and the web GUI.
https://www.tecmint.com/run-deepseek-locally-on-linux/

How come I have never noticed this thread of yours?

I'm not much of an expert in this field, nor I have sufficient experience like you do, but I have a curious mind to try out new things, experiment on them and learn.

Currently, I'm doing experiments with different LLM models locally through Ollama, and LM studio. Mostly it works, but when a large enough model is used, the responses are very slow, it becomes a nightmare, even for my high end device, it struggles to keep up with it  (p.s: RTX 5060Ti 16GB, 32GB RAM, R9 7900X).

The maximum I used with was 32 billion parameters, which is the deepseek r1. It's slow but still usable.

[Vod]

How come I have never noticed this thread of yours?

Cause I locked it for some reason and it became irrelevant.  I'm a year past playing with the LLM models - I'm now deeply reliant on IDEs integrated with AI.  The AI can write difficult code and I use the tools on (Visual code for example) for troubleshooting.  I did upgrade my home server with a GPU, but now its too busy showing me movies to code. 

[Vod]

My workstation became incredibly slow and I couldn't see a reason why - had to reinstall OS yesterday and today I'm reinstalling my apps.   I think it's because I had the antigravity agent as well as multiple Kimi agents both registered on Visual Studio at the same time.  Probably best to uninstall one before installing others.